[PATCH RFC idle 2/3] arm: Avoid invoking RCU when CPU is idle

[Rob Herring]

On 02/02/2012 05:03 PM, Steven Rostedt wrote:
> On Thu, 2012-02-02 at 16:49 -0600, Rob Herring wrote:
>> On 02/02/2012 04:20 PM, Kevin Hilman wrote:
>>> "Paul E. McKenney" <paulmck at linux.vnet.ibm.com> writes:
>>>
>>> [...]
>>>
>>>>>> The two options I see are:
>>>>>>
>>>>>> 1.	Rip tracing out of the inner idle loops and everything that
>>>>>> 	they invoke.
>>>>>
>>>>> What I suggested above.  But as I said I know sh*t about that tracing 
>>>>> implementation so that's an easy suggestion for me to make.
>>>>
>>>> Works for me as well.  ;-)
>>>
>>> While I must admit not having a better suggestion, I for one would vote
>>> strongly against removing tracing from the idle path.
>>>
>>> Being a PM developer and maintainer, much of the code I work on and
>>> maintain happens to be run in the bowels of the idle path.  Not having
>>> the ability to trace this code would be a major step backwards IMO.
>>
>> How is it a step backwards if it is already broken. Obviously you
>> haven't actually used any tracing here because it doesn't work right
>> with things as is.
>>
>> What exactly do you want to trace at this level. By the point you are in
>> this code, the path is somewhat known and problems you have are likely
>> h/w issues. If you are trying to go thru a very precise sequence of
>> saving cpu state and flushing caches, you don't want calls out to
>> tracing code that could very easily change the behavior.
>>
>> It would be nice to understand a real example of how to hit this
>> problem. Why type of tracing? Are there other known examples?
> 
> Well, these are the tracepoints in question:
> 
> trace_power_start(), trace_cpu_idle(), and trace_power_end()
> 
> But this is only used by powertop, and I'm sure nobody would mind if you
> rip these tracepoints out, causing powertop to no longer work. (that's
> why we have 4 bytes per tracepoint wasting space in every trace event).
> 
> Anyway, one answer is (and I was talking with Paul about this on IRC) is
> that we can create a special "TRACE_EVENT_IDLE()" that will explicitly
> call "rcu_idle_exit/enter()" as it expects to be called with it off.
> 
> This should solve most issues I believe.

Seems like these could all be placed common cpuidle code. If you're not
using cpuidle, then you probably aren't doing much in the default idle
code. You would still need the wrappers within the cpuidle code or push
rcu_idle_exit/enter down into a common default idle (which I think we
now have with Nico's clean-up).

Rob